Face Recognition and Identification Using Deep

Learning
2023 Third International Conference on Advances in Electrical, Computing, Communication and Sustainable Technologies (ICAECT) | 978-1-6654-9400-7/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICAECT57570.2023.10118154

Vishwanatha C R V Asha Binju Saju

Department of MCA, New Horizon Department of MCA, New Horizon Department of MCA, New Horizon
College of Engneering, Bengaluru, College of Engineering, Bengaluru, College of Engineering, Bengaluru,
India India India
vishwanathcr@gmail.com asha.gurudath@gmail.com binjusaj@gmail.com

Suma N Talapa Reddy Mrudhula Reddy Sumanth K M

Department of MCA, New Horizon Department of MCA, New Horizon Department of MCA, New Horizon
College of Engineering, Bengaluru, College of Engineering, Bengaluru, College of Engineering, Bengaluru,
India India India
leenasuma6@gmail.com mrudhulareddyt @gmail.com sumanthkm99@gmail.com

Abstract— Face recognition is an unexpectedly growing and problems in computer vision. Before face recognition
extensively carried out component of biometric technologies. techniques can be employed, the technology connected to
Its programs are broad, starting from regulation enforcement face detection is used to find, pinpoint, and extract the face
to customer programs, and enterprise performance and region from the image background. This article will teach
tracking answers. The latest creation of affordable, effective you about:
GPUs and the introduction of massive face databases has
drawn studies consciousness usually at the improvement of an
Face recognition history Deep learning and face
increasing number of deep neural networks designed for all recognition How Working face recognition Face recognition
components identification tasks, starting from detection and applications Deep Face Recognition History Face
pre-processing to characteristic illustration and category in recognition became increasingly popular in the early 1990s
identity verification. Convolutional neural networks can be after the historical Eigenface method was introduced.
used to extract various features from images. A fairly Holistic strategies dominated the face recognition
straightforward method for facial recognition is employed community throughout the 1990s and 2000s. Certain
here. Here, it's crucial to take a deep neural network and distribution assumptions, including linear subspace,
generate a set of digits representing a face (called a face code). manifold, and sparse representation, are used by holistic
The network should produce similar results for both
photographs if you pass two separate images of the same
techniques to determine the low-dimensional representation.
person. But it should produce different results if you give it two The issue with holistic approaches is that they ignore
different photographs of two different persons. The algorithm uncontrollable facial changes that depart from their
need to return very different output for one image. This preconceived notions. In the early 2000s, local feature-based
implies that a neural network must be trained to automatically face recognition was created as a result of this. Local-
recognize various facial features and determine figures based feature-based face recognition and learning-based local
on them. It is possible to imagine that a neural network's descriptors were first introduced in the early 2000s and
output identifies a specific person's face. Passing different 2010s. Gabor filters and Local Binary Pattern (LBP), along
images of the same person, the neural network's output will with their multilevel and high-dimensional extensions, were
resemble one another/near, but passing images of different
people, the output will be very different. This paper presents a
used to achieve face recognition. Working face recognition
technique for recognition of face using the concept of deep Face recognition applications Deep Face Recognition
learning. History Face recognition became increasingly popular in the
early 1990s after the historical Eigenface method was
Keywords— Face Recognition, Deep Learning, Convolutional introduced.
neural network, OpenCV, Algorithm. Holistic strategies dominated the face recognition
community throughout the 1990s and 2000s. Certain
I. INTRODUCTION distribution assumptions, including linear subspace,
Face recognition is a technique for using a person's face manifold, and sparse representation, are used by holistic
to identify or confirm their identification. It is one of the techniques to determine the low-dimensional representation.
most significant applications of computer vision that has The issue with holistic approaches is that they ignore
significant commercial appeal. Deep learning-based uncontrollable facial changes that depart from their
techniques have substantially improved face recognition preconceived notions. In the early 2000s, local feature-based
technology recently. Due to its vast range of applications in face recognition was created as a result of this. Local-
surveillance, law enforcement, biometrics, marketing, and feature-based face recognition and learning-based local
many other fields, face identification in static images and descriptors were first introduced in the early 2000s and
video sequences recorded in unrestricted recording 2010s. Gabor filters and Local Binary Pattern (LBP), along
conditions is one of the most extensively researched with their multilevel and high-dimensional extensions, were

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on October 14,2024 at 20:00:07 UTC from IEEE Xplore. Restrictions apply.
 used to achieve face recognition. As a result, the LFW recognition in unconstrained circumstances. Additionally, It
(Labelled Face in-the-Wild) performance continuously should be mentioned that cutting-edge correctness-achieving
increased from about 60% to above 97% as the solutions are significantly constrained by their reliance
representation pipelines became deeper. Face recognition utilizing advanced GPUs and sizable databases, indicating
milestones in history. Holistic approaches were common in that there is still a sufficient prerequisite for enquiry into
the 1990s and 2000s before local feature-based techniques additional conventional, manually crafted feature
were established in the 2000s and 2010s. Deep learning- representations. Therefore, lowering the excessive
based strategies have dominated since 2014. the source computing cost must be the main goal of future study.
Deep Learning and Face Recognition Numerous deep When there are obstructions between the subject's face
learning techniques have been proposed, with deep and the camera, normal face images are inadequate for IR-
convolutional neural networks (CNN) receiving the most based FR, according to study [4]. Researchers can now
attention in the face recognition field. Since 2014, deep concentrate on Infrared (IR) face photos to solve this
learning technology has transformed the face recognition problem. A multi-dimensional imaging system is provided
research environment thanks to the innovations of Deep by IR images. In unfavourable situations Object lighting,
Face and DeepID approaches. Since then, deep face altered facial expressions, disguised faces, dark surrounds,
recognition algorithms have advanced, using hierarchical and three-dimensional imagery technology is employed to
architecture to develop discriminative face representation. provide higher precise findings. issue, researchers may now
Modern performance was much enhanced, and several concentrate on Infrared (IR) face photos. A multi-
successful real-world applications were promoted. Deep dimensional imaging system is provided by IR images. In
learning uses many processing layers to learn unfavourable situations such techniques as facial cloaking,
representations of input with different levels of feature object illumination, emotion alterations, and darkness, the
extraction. multi-dimensional imaging technology is employed to
provide higher precise environments.
II. RELATED WORK Tuba [5] used Convolutional Neural Networks (CNN) to
The notion of designing and creating by using Python and build a system that can recognize students' moods from their
OpenCV, a face recognition system is being developed. faces. On the 2013 dataset, they classified seven different
described in the Teoh’s [1] paper. Deep learning can be used emotions using facial expressions with an accuracy of
to do facial recognition, and given its high degree of 74.41% and validation accuracy of 77.00%.
precision, it seems to be an appropriate method. Results Deep learning techniques have been evolved in the recent
from experiments demonstrate the precision of the times in multiple fields including agricultural sector where
suggested facial recognition system. The proposed facial they are used for detection of Locust Outbreaks in
recognition system's accurate and false recognition can be Agricultural Harvesting [6]. The MTCNN method proposed
seen in action. 17 out of 20 recognitions for a first person by T. D. Tithy et al. [7] was for the security camera
are accurate. The system's recognition of a person's generated video frames. The method employed deep
identification in a photo matches that person's actual identity, learning approach and the model was trained using
making the assertion true. As a result, it is possible to architecture related to the Res-Net. In this method an alarm
determine the system's accuracy. will be generated if the system fails to recognize any face.
Singh discussed using a special combination of Ayush kumar et al. [8] used a Pragmatic Approach which
techniques in the work [2], including Deep recurrent was based on Novel Deep Learning and employed a hybrid
learning, SMQT (Successive Mean Quantization Transform), based technique.
and K-nearest neighbour’s (KNN) are all examples of
machine learning techniques. The method chosen in this III. METHODOLOGY
instance makes intelligent use of factors like noise, dim
lighting, haziness, and a tiny slant in the photograph. The A. Face Recognition using Deep Learning
research will point the way to effective facial data Machine Learning Engineering, Deep Learning is an area
recognition while taking into account various image effects. of Artificial Intelligence (AI) that aims to bring human brain
These techniques were chosen because they resemble neural functions into data processing machines. The activity of the
networks, as well as that basic machine learning techniques human brain serves as the basis pertaining to deep learning,
cannot manage the purpose of a brief amount of time with sometimes referred to as deep neural learning. Data
good precision. According on this research’s findings, a processing is an important field, and deep learning helps to
video database is 98.72% effective. Although this research process large amounts of data using identified and verified
could accurately identify the video, it was unable to identify patterns established by the human brain. A revolutionary
the people's names. machine learning technique, deep learning has enabled the
The method by M. H. Robin et al. compares and contrasts field of technology to evolve in many ways.
the databases that are currently accessible and relates One of the uses of deep learning can be seen in image
benchmarks [3]. It examined solutions intended to overcome colouring. You have read correctly. Deep learning now
these constraints and highlighted unresolved concerns as it makes it possible to add colour to black and white photos.
outlined the drawbacks of cutting-edge methodologies. Unlike traditional techniques that rely on people to do this
Despite tremendous improvements in representation work manually, the process of colouring images and adding
accuracy brought on by the non-linearity of deep feature vibrant colours to black and white photos can be done by
representations, we can categorically assert that no known machines in script. nowadays. Working on models of the
ideal facial feature is sufficiently resilient for face human brain and neural networks, deep learning

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on October 14,2024 at 20:00:07 UTC from IEEE Xplore. Restrictions apply.
 applications in image processing allow computers to learn the dlib library, we estimate the 68 (x, y) coordinate
colours and set up algorithms that allow us to colour. positions mapping the facial structure of the face. Images
may be operated on with Cv2. This step loads all pretrained
B.Convolutional Neural Network (CNN) packages. You need these four pretrained files to work with
Deep learning employs sophisticated algorithms and it. Now initialize two lists. One for face and one for name.
artificial neural networks to analyses large amounts of data. Store the 128-dimensional face codes obtained from the
learn from experience and to be able to similar to how the name list's associated labels and the face list's face encoder,
human brain classifies and recognizes information. Train which uses all training pictures. Also, in this case define a
your machine / computer to regarding deep learning. route that contains a model of training images. I load this
Artificial neural networks of the convolutional variety are image with cv2, to locate faces in the picture, change it to
often used to identify objects and images as well as to grayscale, and then feed the photograph to the cv2.dnn
classify them. Hence, deep learning uses CNNs for object model. Afterward, loop through each detection that the cv2-
recognition in images., localization and segmentation jobs dnn model offers. Here, 0.7 has been chosen as the
for computer vision, video analytics for self-driving vehicle confidence level When using Face Aligner to align the faces,
obstacle detection, and image processing issues etc. increase. detections with a confidence value higher than 0.7 are
CNNs are very popular in deep learning because they play a utilized as face pictures. We then extract the passing the
key role in these rapidly growing new areas. aligned faces and landmarks to a face encoder after
extracting landmarks from the aligned faces, a
trained Resnet model. The pre-trained model provides a
(128,1)-dimensional encoding for all images and appends
the labelled output to the face and name lists respectively.
After getting all the image encodings and adding them to a
list, convert this list to a numpy array and store it in. npy
format to the hard drive. By doing this, I was able to store
the list on my hard drive and avoid having to build it each
time I do facial recognition. You can import your list at any
time. A list is called a pattern. Currently, your model has
been built and stored.
Performance Analysis: Three parameters are used to
assess the facial recognition system's performance. Training
data, pose variation, and picture illumination variation are
Fig. 3.1 Block diagram of CNN
them.

D. Face Recognition using OpenCV

C.Face Recognition Implementation Steps It acts like a Python layer over OpenCV’s C++ code.
Dataset collection: A Dataset is a collection of data Because the core code is written in C/C++, OpenCV-Python
arranged in a specific order. A dataset can contain anything is not only fast, it is also easy to develop and deploy (thanks
from a set of arrays to a database table. A tabular record can to a front-end Python wrapper). This makes it an awesome
be thought of as a database table or matrix, with each option to run applications that need lots of computing power.
column corresponding to a particular variable and each row Steps in Face recognition: Face recognition is frequently
corresponding to a field in the record. stated as the first stage of a four-step process that includes
Kaggle is one of the best sources for providing data sets face recognition, face alignment, feature extraction, and
for data scientists and machine learning. Users can easily finally facial recognition.
search, download, and publish records. It also provides
opportunities to collaborate with other machine learning 1. Face identification: Find a face or faces in the picture,
engineers to solve difficult tasks related to data science. and then draw a bounding box around them.
Kaggle provides high-quality datasets in a variety of formats 2.Face placement: Faces should be normalized to fit
that are easy to find and download. databases like geometries and photometrics.
Import the data set: The next step is to load the data set
into Collab. To import a dataset, we first need to import 3. Mining features: Mining of facial for the recognition
some of the required libraries. Once that's done, let's import job, characteristics might be employed.
the data currently in Google Drive using the code. 4. Face identification: Compare a picture to one or more
Data Pre-processing: After importing the library and data, known pictures in the database you have generated.
let's proceed to data pre-processing.
Unlike in the past, a system may integrate any or all of
Images exist in many different formats, with natural,
theprocesses into a single process instead of having a
spurious, greyscale, etc., and they should be considered and
distinct module or programmed for each phase.
normalized before input to the neural network.
Model Building: Downloading the dataset is the initial
step before running our code. As follows many facial
datasets available online. After You may start with the code
after obtaining the dataset. We’ve loaded a few Python
packages here. Use of dlib in face feature recognition. Using
face analysis experts feature recognition points from

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on October 14,2024 at 20:00:07 UTC from IEEE Xplore. Restrictions apply.
 faced Considering the issue that face to face in several
directions looks completely variant from the computer.
The fundamental premise is that we will discover 68
particular spots (referred to as landmarks) that exist on
every face, such as the tip of the chin, the inside border of
each brow, and so forth. The next step is to train a machine
learning system to be able to locate those 68 precise
locations on any given face.

Fig. 3.2Face Recognition Chart

5. Implementation:Next Using Python and OpenCV, we

will put facial recognition into practise. Let's first examine
the libraries we will require and their installation procedures
of OpenCV, dlib and Face Recognition.
A video and image processing package called OpenCV is
used for tasks including face recognition, reading licence
plates, picture editing, improved robotic vision, and more.
The dlib library includes the "Metric Deep Learning"
implementation utilised to construct our face integrations
employed in the recognition process itself. We will be
utilising the face recognition library in our code because it is Fig. 4.1 Detecting faces
so simple to utilise.
6. Extract characteristics from Face: A dataset is first We have reached the crux of the issue: genuinely
required; you may even make one yourself. Just be careful distinguishing between faces. Things start to become
to group all of your photos into folders, with one folder extremely intriguing at this point. Comparing the unknown
holding all of a certain person's pictures. face, we discovered in Step 2 straight to all the images we
have of people who have previously been tagged is the
7. Encode our face image: Data and computing power. simplest method for face recognition. It must be the same
However, after training, the network can produce metrics for individual if we discover a previously tagged face that
any aspect, even ones it has never seen before! As a result, closely resembles our unidentified face. When a face is
you only need to do this step once. Fortunately for us, the found, the application creates a rectangle around it. If more
excellent people at Open-faced have already completed this than one face is found, more than one rectangleis there.
task and have made a number of trained networks available
for live usage. Therefore, all we had to do to obtain 128
measures for each face was to feed their trained network our
face photos.

IV. RESULT AND DISCUSSION

Deep Learning is a task that requires a lot of data, and in
the case of face recognition, it is not always possible to
process large amounts of data. As such, advances have
improved the usability and viability of face recognition by
one-shot learning. We developed a Python module that was
even simpler to use in order to make it even simpler.
Skipped all steps in Downloading Support Files and
Configuring Support Files. Just install with the python
module, you can begin face recognition and identification. Fig. 4.2 Comparing face images
The first step is to download the data set so that you can
start executing your code. Including many facial datasets This final step is actually the easiest in this entire process.
available online. All we need to do is to find the individual from our known
Open CV:The initial action we take is to recognize faces people database that has the closest measurements from our
in images (photos) or video streams. Great camera testing images. All we need to do is train a classifier which
functionality includes face recognition. Before taking a is capable of taking measurements from a new test image
photo, the camera may make sure that all of the faces are in and telling us which known people are the closest matches.
focus by having the ability to automatically choose faces. It takes a few milliseconds to train that classifier.
Since we don't require color information to detect faces, we Performance analysis of CNN: The emergence of faster
will begin by turning our image black and white in order to CPUs and higher-resolution cameras has spurred the search
find faces in it. After that, we will examine our image pixel to develop facial recognition systems for different
by pixel. We wish to take into account the pixels that applications. Facial recognition systems either employ off-
immediately around each pixel. Well, we have the faces in line data or live input, depending on the application. In this
our photograph were isolated. However, at this time, are paper, we present a design and evaluation of a real-time

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on October 14,2024 at 20:00:07 UTC from IEEE Xplore. Restrictions apply.
 facial recognition system using convolutional neural V. CONCLUSIONS
networks (CNNs). Today, facial recognition is used in many industries
around the world. Here are his seven best uses. Unlocking
phones, smarter advertising, finding missing persons,
protecting law enforcement, identifying people on social
media, tracking student attendance, and forensics. Advances
in security and surveillance have changed how data is
collected, how activities are driven, and how data can be
best afterwards utilized. Security systems can be as
straightforward as video cameras or as complicated as
biometric systems in order to detect, recognize, and record
breaches.
We have used deep learning in this study approach to
construct a face recognition system and identification
system. We describe the Employing a CNN technique to
facial recognition, this system's whole development process
Fig. 4.3 Face Recognition output started with training data. It is proven that images and real-
time videos can be accurately recognized by training a
classifier with a large number of face images. Facial
identificationis an element of computer vision. Facial
identification is employed to identify aindividual in a
biometric authentication process based on facial images.
People are identified by their biological characteristics. The
human eye may quickly identify individuals merely by
glancing at them, but there is a limit to how much the
human eye can focus. Therefore, a computer-assisted
method was invented to perform face recognition. Facial
recognition involves automatically recognizing and
verifying people from images or videos. The obtained
outputs justifies that the method is able to detect or
Fig. 4.4 Face Recognition output recognise the face successfully. The time taken by the
algorithm is around 360 to 390 milliseconds which is quite
Initial evaluation of the proposed design is performed optimal considering the processing time.
using AT&Ts stock data set, which is then extended to real- REFERENCES
time system design. Details of tuning the CNN parameters
[1] M. Arsenovic, S. Sladojevic, A. Anderla and D. Stefanovic,
for evaluating and improving detection accuracy of the "FaceTime — Deep learning based face recognition attendance
proposed system are also reported. A systematic method for system," 2017 IEEE 15th International Symposium on Intelligent
tuning parameters for improving system performance is also Systems and Informatics (SISY), 2017, pp. 000053-000058, doi:
proposed. 10.1109/SISY.2017.8080587.
[2] M. T. H. Fuad et al., "Recent Advances in Deep Learning
Techniques for Face Recognition," in IEEE Access, vol. 9, pp.
99112-99142, 2021, doi: 10.1109/ACCESS.2021.3096136.
[3] M. H. Robin, M. M. Ur Rahman, A. M. Taief and Q. Nahar Eity,
"Improvement of Face and Eye Detection Performance by Using
Multi-task Cascaded Convolutional Networks," 2020 IEEE Region
10 Symposium (TENSYMP), 2020, pp. 977-980, doi:
10.1109/TENSYMP50017.2020.9230756.
[4] A. Singh, "Face Detection using Deep Recurrent Learning and
SMQT Technique," 2020 International Conference on Electronics
and Sustainable Communication Systems (ICESC), 2020, pp. 17-22,
doi: 10.1109/ICESC48915.2020.9155996.
[5] Teoh, Kh, Rc Ismail, Szm Naziri, R Hussin, Mnm Isa, and Mssm
Basir, ‘Face Recognition and Identification Using Deep Learning
Approach’, Journal of Physics: Conference Series, 1755.1 (2021)
[6] N. Krishnamoorthy, D. R. Suresh, D. Mohanapriya, D. A. Prasad, D.
R. Krishnamoorthy, and D. R. Thiagarajan, “Utilisation of Deep
Learning to Exploit Locust Outbreaks in Agricultural Harvesting,”
vol. 20, no. 10, p. 8, 2022.
[7] T. D. Tithy, S. Chakraborty, R. Islam and A. Aziz, "A Deep
Learning based Approach for Real Time Face Recognition System,"
2021 International Conference on Electronics, Communications and
Information Technology (ICECIT), 2021, pp. 1-4, doi:
10.1109/ICECIT54077.2021.9641476.
[8] A. Kumar, P. Kumari and N. Suresh kumar, "A Pragmatic Approach
to Face Recognition using a Novel Deep Learning Algorithm," 2021
International Conference on Advance Computing and Innovative
Technologies in Engineering (ICACITE), 2021, pp. 806-810, doi:
10.1109/ICACITE51222.2021.9404697.
Fig. 4.5 Graph for face recognition

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on October 14,2024 at 20:00:07 UTC from IEEE Xplore. Restrictions apply.